Assess · Healthcheck
Spec-to-implementation drift audit
Spec-to-implementation drift audit. Where is your team losing intent?
1 week
Discuss this starting pointDevelopment
Coherence
The intent your team wrote down on Monday and the code that shipped on Friday are rarely the same thing. AI coding assistance makes the gap wider, faster.
Vibe-coding is the polite name for hands-off AI coding. On the gradient from heavily-reviewed AI assistance to fully hands-off generation, drift between intent and implementation goes up the more hands-off you get. Solo-risky; team-critical.
Intent Coherence is the discipline of keeping spec and code aligned while you ship. Hygiene, not ritual: a specification refinement loop, drift detection, and evaluation coverage as the mechanism that confirms the two still match.
How it works
The practice
This pair sits inside Operational Excellence in the VOF. Coherence is its build-time half — keeping spec and code aligned while you're shipping. GenAIOps is its run-time half — keeping the live system honest once it's out there. This page is the build-time half.
Layer 1 — Intent Coherence (the discipline)
Intent Coherence has three loops. Together they form the hygiene that keeps AI-assisted development from drifting.
Specification refinement. The spec is a living artefact, not a one-shot. As implementation reveals what the spec didn't pin down, the spec gets refined and the code follows.
Drift detection. Something — review, eval, diff — flags when code and intent part company. Continuous, not annual. The earlier you catch a drift, the cheaper it is to close.
Eval coverage as the coherence mechanism. The eval suite is the assertion that intent and implementation still agree. If the spec changes and the evals don't, you've lost coherence — you just haven't measured it yet.
Layer 2 — Lode Coding (the practice — by FJ Zeit)
Lode Coding is FJ Zeit's work — the specific practice Viewyonder recommends and uses inside Intent Coherence. The discipline is Steve's framing; the practice is FJ Zeit's.
A lode is a folder of markdown files maintained by the AI agent itself — summary, terminology, practices, lode-map, domain-specific files. The agent creates and maintains the lode through conversational orchestration. Named after a vein of valuable ore.
Lode Coding by FJ Zeit.
Read the practice →
fjzeit.github.io/coding/lode-coding-basics
Browse the lode repo →
fjzeit.github.io/lode
Buy a starting point
Three ways to start
Plan · Plan
Intent Coherence playbook + Lode design
Intent Coherence playbook + Lode design for your stack.
2–3 weeks
Discuss this starting pointExecute · Execute
Implement tooling, embed in workflow
Implement Coherence tooling and lode, embed in workflow, train team.
4–12 weeks
Discuss this starting pointTake it with you